Macro-Founded Machine Learning Models for Power Market Price Trend Detection

The paper explores the development of a trading strategy in the power market using Artificial Intelligence algorithms to exploit potential information from a large set of indicators. The first part of the document relies on feature construction, analyzing key market factors such as gas storage capacity, withdrawals, and injections to define macroeconomic regimes. The second part introduces additional complexity either testing the granger-causal relationship between gas flow and futures contract returns, either considering temperature-related variables and weather shocks features to expand the training input space of the Machine Learning (ML) models. Finally, the study presents an out-of-sample simulation, comparing ML model performances with a naïve-long only strategy.